Substitution method of single-material cigarette in cigarette leaf group formula and application of substitution method

Abstract

The invention relates to a replacement method for single-material cigarettes in a cigarette leaf formula and application of the replacement method. According to the bipartite graph path selection aggregation strategy-based tobacco leaf group single-material tobacco replacement method, a bipartite graph data model of tobacco leaf group formula-single-material tobacco information is constructed, and the path selection aggregation strategy is adopted; according to the method, the to-be-replaced single-material cigarettes are globally expressed in all-leaf-group formula data from two dimensions of attributes of leaf groups to which the to-be-replaced single-material cigarettes belong and characteristics of the to-be-replaced single-material cigarettes belong, recommendation of the replaced single-material cigarettes is completed from the angle of leaf group adaptation not only aiming at characteristic parameters of target single-material cigarettes, the uncertainty of traditional single-material cigarette replacement depending on artificial sensory evaluation can be avoided, and the recommendation efficiency is improved. The scientificity and the efficiency of the replacement selection work of the tobacco leaf group formula single-material tobacco are improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of cigarette leaf group formula design and maintenance, and relates to a method for substituting single-leaf tobacco in a cigarette leaf group formula and its application. Background Art

[0002] In the tobacco industry, the cigarette leaf group formula directly determines the quality and sales volume of cigarette products. However, since tobacco leaves are planting products, they are easily affected by the climate of the production area, resulting in problems such as decreased yield and unstable quality. For cigarette products sold on the market, it is necessary to replace the single-leaf tobacco in the formula according to the tobacco leaf production situation of the current year to ensure product quality. Therefore, how to select the replacement of single-leaf tobacco is a key issue in the tobacco industry, which directly determines the quality of cigarette products. The replacement selection of single-leaf tobacco is a challenging task, which not only requires a high degree of similarity in physical and chemical properties between the old and new single-leaf tobacco raw materials, but also needs to evaluate whether these changes are suitable for the overall characteristics of the final cigarette product. At present, the task of replacing single-leaf tobacco mainly relies on the sensory evaluation of tasting personnel for manual selection. However, this method places extremely high requirements on the professional knowledge and practical operation experience of technicians, and at the same time, this decision-making method based entirely on personal experience inevitably has subjective uncertainty. Therefore, developing a method for replacing single-leaf tobacco in a cigarette leaf group formula based on non-artificial experience has gradually become one of the research hotspots in this field.

[0003] For example, CN115868656A discloses a method for simulating and formulating a cigarette leaf group formula based on tobacco leaf substitution, including: constructing a tobacco leaf data information library; performing linear discriminant analysis classification training based on the tobacco leaf data information library to obtain a tobacco leaf class label prediction model; selecting a target leaf group formula Q to be simulated and formulated, predicting the class label of each type of tobacco leaf in Q respectively to obtain the corresponding class label; obtaining the optimal replacement tobacco leaf and the sub-optimal replacement tobacco leaf of each type of tobacco leaf in Q; for each type of tobacco leaf in Q, using the corresponding optimal replacement tobacco leaf for replacement, and when the optimal replacement tobacco leaf is also a component in Q, then selecting the sub-optimal replacement tobacco leaf for replacement to complete the generation of formula P; optimizing the ratio of formula P. By simulating and formulating the existing cigarette leaf group formula, a leaf group formula list with a similar style is generated, achieving the design of a cigarette leaf group formula based on non-artificial experience and improving the design efficiency of the cigarette leaf group formula.

[0004] In summary, developing a new and efficient method for substituting single-leaf tobacco in a cigarette leaf group formula and expanding the design tools for the cigarette leaf group formula is of great significance to the tobacco industry. Summary of the Invention

[0005] Aiming at the deficiencies of the prior art and the actual needs, the present invention provides a method for substituting single-leaf tobacco in a cigarette leaf group formula and its application, developing a new method for recommending the replacement of single-leaf tobacco in a cigarette leaf group, and improving the scientificity and efficiency of the work of replacing single-leaf tobacco in the cigarette leaf group formula.

[0006] To achieve this goal, the present invention adopts the following technical solutions:

[0007] In a first aspect, the present invention provides a method for substituting single tobacco leaves in a cigarette leaf group formula, and the method includes the following steps:

[0008] (1) Obtain the information of each cigarette leaf group formula, where the information includes the attribute information of the cigarette leaf group formula and the characteristic information of all single tobacco leaves therein;

[0009] (2) Based on a bipartite graph, model each cigarette leaf group formula and the single tobacco leaf information therein. The nodes in the bipartite graph are the information of each cigarette leaf group formula or the single tobacco leaves therein;

[0010] (3) Integrate the attribute information of the target cigarette leaf group formula in the bipartite graph and the characteristic information of all single tobacco leaves therein as the integrated feature of the target cigarette leaf group formula;

[0011] (4) Select the cigarette leaf group formula nodes adjacent to the target cigarette leaf group formula and perform step (3) to obtain the integrated features of the adjacent cigarette leaf group formulas, where the adjacent cigarette leaf group formulas refer to all cigarette leaf group formula nodes that use the same single tobacco leaves as the target cigarette leaf group formula;

[0012] (5) Calculate the similarity of the integrated features between the adjacent cigarette leaf group formulas and the target cigarette leaf group formula, and select the top K adjacent cigarette leaf group formulas with the highest similarity to the target cigarette leaf group formula;

[0013] (6) Select the single tobacco leaf nodes in the K adjacent cigarette leaf group formulas obtained in step (5) as the alternative available substitute single tobacco leaves;

[0014] (7) Construct a global correlation subgraph of the node of the single tobacco leaf to be substituted. Starting from the node of the single tobacco leaf to be substituted to the node of the alternative substitute single tobacco leaf as the end point, explore all path sequences in the global correlation subgraph respectively for global feature integration;

[0015] (8) Starting from all alternative available substitute single tobacco leaf nodes in the global correlation subgraph constructed in step (7) to the node of the single tobacco leaf to be substituted as the end point, explore all paths in the subgraph and then perform global feature integration respectively;

[0016] (9) Calculate the similarity of the positive and negative global integration features between all alternative substitute single tobacco leaf nodes and the target single tobacco leaf node to be substituted, and select the top M single tobacco leaves with the highest similarity to the target single tobacco leaf node to be substituted as the final recommended substitute single tobacco leaves.

[0017] The present invention models the leaf group and single flue-cured tobacco information based on a bipartite graph, and uses a path selection and aggregation strategy to globally express single flue-cured tobacco in terms of the attributes of its affiliated leaf group and its own characteristics in the overall leaf group formula data. It not only targets the characteristic parameters of the target single flue-cured tobacco, but also completes the recommendation of substitute single flue-cured tobacco from the perspective of leaf group adaptation, which can avoid the uncertainty of traditional single flue-cured tobacco recommendation relying on manual sensory evaluation and improve the scientificity and efficiency of the single flue-cured tobacco substitution selection work for leaf group formulas.

[0018] Preferably, the information in step (1) specifically includes the grade, type, price of the cigarette leaf group formula, and the aroma, fragrance type, smoke, aroma, and taste characteristic parameters of each single flue-cured tobacco.

[0019] Preferably, in the bipartite graph structure in step (2), the cigarette leaf group formula nodes are only connected to the single flue-cured tobacco nodes, and the single flue-cured tobacco nodes are also only connected to the cigarette leaf group formula nodes. There is no connection between the cigarette leaf group formulas and the cigarette leaf group formula nodes, nor between the single flue-cured tobaccos and the single flue-cured tobacco nodes.

[0020] Preferably, step (3) specifically includes: taking the single flue-cured tobacco to be substituted in the target cigarette leaf group formula as the starting point, and fusing the attribute information of the cigarette leaf group formulas of its 1st-order neighbors and 2nd-order neighbors in the bipartite graph structure and the characteristic information of all the single flue-cured tobaccos therein.

[0021] Preferably, step (5) specifically includes: using a similarity measurement calculation method, comparing the distances between the fusion feature vectors of all neighboring cigarette leaf group formulas and the target cigarette leaf group formula as the similarity evaluation index, and selecting the K cigarette leaf group formulas with the closest distance to the target cigarette leaf group formula.

[0022] Preferably, step (7) specifically includes: taking the single flue-cured tobacco node to be substituted as the starting point, its 1st-order neighbor which is the affiliated cigarette leaf group formula, the 2nd-order neighbor which is the other single flue-cured tobaccos in the affiliated cigarette leaf group formula, the 3rd-order neighbor which is the K neighboring cigarette leaf group formulas, and the 4th-order neighbor which is the single flue-cured tobaccos in the K neighboring cigarette leaf group formulas. The above four parts constitute the global relevant subgraph of the single flue-cured tobacco node to be substituted. Then, starting from the single flue-cured tobacco node to be substituted to the 4th-order node as the end point, all paths in the global subgraph are explored for feature fusion respectively.

[0023] Preferably, step (8) specifically includes: in the global relevant subgraph obtained in step (7), starting from the 4th-order neighbor node to the single flue-cured tobacco node to be substituted as the end point, all paths in the global subgraph are explored in reverse for global feature fusion respectively.

[0024] Preferably, step (9) specifically includes: using a similarity measurement calculation method, comparing the distances between the global fusion feature vectors of all alternative substitute single flue-cured tobaccos and the target single flue-cured tobacco node to be substituted as the similarity evaluation index, and selecting the M single flue-cured tobaccos with the closest distance to the target cigarette leaf group formula.

[0025] Preferably, step (1) further includes obtaining the characteristic vectors of the grade, type, price of the cigarette leaf group formula, and the flavor, smoke, aroma, and taste of each single tobacco leaf, and constructing a data set, wherein the characteristic vector of the cigarette leaf group formula includes the grade, type, and price of the cigarette leaf group formula, and the characteristic vector of the single tobacco leaf includes flavor, degree, concentration, strength, aroma, aroma quantity, permeability, miscellaneous odor, irritation, dryness, and aftertaste.

[0026] Preferably, step (1) further includes obtaining the percentage of the usage of each single tobacco leaf in the cigarette leaf group formula. For reference.

[0027] Preferably, step (2) specifically includes the following steps:

[0028] (2.1) According to the belonging relationship between each cigarette leaf group formula and different single tobacco leaves, construct a graph structure G = (V, E), where V = {v l 1, v l 2,..., v l n; v c 1, v c 2,..., v c m} refers to the nodes in the graph, that is, all cigarette leaf group formulas and single tobacco leaves; E refers to the edges between the nodes, that is, the connection between the cigarette leaf group formula and the single tobacco leaf. If the two are in a belonging relationship, there is a corresponding edge between the cigarette leaf group formula node v l and the single tobacco leaf node v c , otherwise there is no such edge;

[0029] (2.2) Calculate the adjacency matrix of the graph where <v l i, v c j> refers to the edge between the nodes v l i and v c j, that is, if there is an edge between the cigarette leaf group formula node v l i and the single tobacco leaf node v c j, then the value of their adjacency matrix is 1, otherwise it is 0. Based on this, the two-dimensional adjacency matrix A[i, j] can be obtained;

[0030] where, for the cigarette leaf group formula node set v l and the single tobacco leaf node set v c , since there is no edge connection between them internally, the values of A[v l i, v l j] and A[v c i, v c j] in the adjacency matrix are both 0, and only between the cigarette leaf group formula node and the single tobacco leaf node is there A[v i i, v lj]=*, when <v l i, v c When <j> exists, *=1, otherwise *=0;

[0031] (2.3) According to the results of steps (2.1)-(2.2), draw a bipartite graph structure diagram of the cigarette leaf group formula - single-flavor cigarette, where v l and v c are the cigarette leaf group formula node and the single-flavor cigarette node respectively. The connection between the two is the edge E. There is no edge inside the cigarette leaf group formula node and the single-flavor cigarette node. The nodes are presented as two relatively independent parts as a whole, that is, the bipartite graph model.

[0032] Preferably, step (3) specifically includes the following steps:

[0033] (3.1) Let the feature vector fc i of single-flavor cigarette i = [aroma type, degree, concentration, strength, aroma, aroma quantity, permeability, off-flavor, irritation, dryness, aftertaste]; the percentage of the usage amount of the single-flavor cigarette in the cigarette leaf group formula is w i , and the feature vector fl j of the cigarette leaf group formula = [cigarette leaf group formula grade, cigarette leaf group formula type, price]; the fusion feature vector of the cigarette leaf group formula is

[0034] (3.2) The calculation formula of the fusion feature vector of the cigarette leaf group formula is: where || represents vector splicing, k is the number of types of single-flavor cigarettes in cigarette leaf group formula j, refers to weighted summation of the parameter of each single-flavor cigarette in the cigarette leaf group formula by the usage amount percentage. The obtained fusion feature vector of the cigarette leaf group formula can be expressed as:

[0035] where aroma type’, degree’, concentration’, strength’, aroma’, aroma quantity’, permeability’, off-flavor’, irritation’, dryness’, aftertaste’ are the weighted fusion values of the index parameters of each single-flavor cigarette in the cigarette leaf group formula.

[0036] Preferably, step (5) specifically includes the following steps:

[0037] (5.1) For the fusion feature vector

[0038] Perform a normalization operation. For the numerical variable parameters price, degree, concentration, strength, aroma, aroma quantity, permeability, off-flavor, irritation, dryness, and aftertaste, the normalization calculation formula is where x is the original value of the variable, x max is the maximum value of the variable, x min is the minimum value of the variable, x norm is the normalized value, and the value range is 0 - 1; convert the cigarette leaf group formula grade, cigarette leaf group formula type, and flavor type into digital variables. Among them, the cigarette leaf group formula grade is divided into Grade A-1 (which can be numbered as the number 1), Grade A-2 (which can be numbered as the number 2), Grade B-1 (which can be numbered as the number 3), and Grade B-2 (which can be numbered as the number 4); the cigarette leaf group formula type is divided into flue-cured type (which can be numbered as the number 1), blended type (which can be numbered as the number 2), flavor type (which can be numbered as the number 3), and cigar type (which can be numbered as the number 4); the flavor type is divided into light flavor type (which can be numbered as the number 1), intermediate flavor type (which can be numbered as the number 3), and strong flavor type (which can be numbered as the number 4); then participate in the subsequent similarity calculation;

[0039] (5.2) Calculate the fusion feature similarity. Adopt a method combining the cosine of the angle and the Euclidean distance. For the normalized numerical variable parameters price, degree, concentration, strength, aroma, aroma quantity, permeability, off-flavor, irritation, dryness, and aftertaste, use the Euclidean distance to calculate:

[0040]

[0041] For the descriptive parameters cigarette leaf group formula grade, cigarette leaf group formula type, and flavor type that are converted into digital variables, use the cosine of the angle to calculate:

[0042]

[0043] where x i and y i are the respective feature parameters in different feature vectors. The final fusion feature similarity of the cigarette leaf group formula is calculated by the formula The smaller this value is, the more similar the fusion features of the two cigarette leaf group formulas are;

[0044] (5.3) Calculate all neighboring cigarette leaf group formulas and the target cigarette leaf group formula between the Sim Select the K neighboring cigarette leaf group formulas with the highest similarity among them.

[0045] Preferably, step (6) specifically includes: starting from the target single cut tobacco, the first-order adjacent nodes are the tobacco leaf group formulas of the cigarettes, the second-order adjacent nodes are the other single cut tobaccos in the same tobacco leaf group formula, the third-order adjacent nodes are the adjacent tobacco leaf group formula nodes, and the fourth-order adjacent nodes are the single cut tobaccos of the adjacent tobacco leaf group formulas. The single cut tobaccos of the adjacent tobacco leaf group formulas are the alternative available substitute single cut tobaccos.

[0046] Preferably, step (7) specifically includes the following steps:

[0047] (7.1) Construct a global relevant subgraph of the single cut tobacco node to be replaced. The global relevant subgraph includes the target single cut tobacco in step (6) as the starting point, the first-order adjacent nodes, the second-order adjacent nodes, the third-order adjacent nodes, and the fourth-order adjacent nodes;

[0048] (7.2) For the global relevant subgraph described in step (7.1), explore all path sequences from the target single cut tobacco node to the K fourth-order adjacent single cut tobacco nodes of the adjacent tobacco leaf group formulas The calculation formula for the information fusion feature vector of each path sequence is: where fc t is the feature vector of the single cut tobacco to be replaced; fc ki is the feature vector of the alternative available substitute single cut tobacco node i in the adjacent tobacco leaf group formula k; is the fusion feature of the cigarette tobacco leaf group formula where the single cut tobacco to be replaced is located; is the fusion feature vector of the adjacent tobacco leaf group formula k, and w k is the weighted value, and the calculation formula is: where is the fusion feature vector of the cigarette tobacco leaf group formula where the single cut tobacco to be replaced is located calculated by using step (5.2) and the fusion feature vector of the adjacent tobacco leaf group formula k the similarity between them.

[0049] Preferably, step (8) specifically includes:

[0050] Explore the path sequence from the alternative available substitute single cut tobacco node to the single cut tobacco node to be replaced The calculation formula for the information fusion feature vector of each path is: where fc t is the feature vector of the single cut tobacco to be replaced; fc ki is the feature vector of the alternative available substitute single cut tobacco node i in the adjacent tobacco leaf group formula k; is the fusion feature of the cigarette tobacco leaf group formula where the single cut tobacco to be replaced is located; is the fusion feature vector of the adjacent tobacco leaf group formula k, and w k is the weighted value, and the calculation formula is: Among them is the fusion feature vector of the cigarette leaf group formula where the single tobacco leaf to be replaced is calculated by step (5.2) and the fusion feature vector of the adjacent cigarette leaf group formula k The similarity between them

[0051] Preferably, step (9) specifically includes: obtaining the forward and reverse path fusion feature vectors fpath from the node t of the single tobacco leaf to be replaced to the node ki of the alternative available single tobacco leaf according to steps (7) and (8) t→ki and fpath ki→t After that, the Manhattan distance is used to calculate the similarity between the two, and the calculation formula is Among them, x i and y i are the respective feature parameters in fpath t→ki and fpath ki→t respectively. The larger the similarity value, the more similar the two are. Select the M alternative single tobacco leaves ki with the highest similarity as the final recommended alternative single tobacco leaves

[0052] In the second aspect, the present invention provides an application of the method for replacing a single tobacco leaf in the cigarette leaf group formula described in the first aspect in designing a cigarette leaf group formula

[0053] Compared with the prior art, the present invention has at least the following beneficial effects

[0054] The present invention designs a method for replacing single tobacco leaves in a leaf group based on a bipartite graph path selection aggregation strategy. By constructing a bipartite graph data model of leaf group formula - single tobacco leaf information and adopting a path selection aggregation strategy, the single tobacco leaf to be replaced is globally expressed in the whole leaf group formula data from two dimensions of the attributes of the leaf group to which it belongs and its own characteristics. Not only for the target single tobacco leaf characteristic parameters, but also from the perspective of leaf group adaptation, the recommended alternative single tobacco leaves are completed, which can avoid the uncertainty of traditional single tobacco leaf recommendation relying on manual sensory evaluation and improve the scientificity and efficiency of the single tobacco leaf replacement selection work in the leaf group formula BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Figure 1 It is a schematic diagram of the bipartite graph structure of the cigarette leaf group - single tobacco leaf

[0056] Figure 2 It is the global relevant sub - graph of the single tobacco leaf A to be replaced DETAILED DESCRIPTION OF THE EMBODIMENTS

[0057] The technical solutions of the present invention will be further described below with reference to the drawings and through specific embodiments. However, the following examples are only simple examples of the present invention and do not represent or limit the scope of the protection of the rights of the present invention. The scope of protection of the present invention shall be subject to the claims

[0058] For those technical details or conditions not specified in the examples, they shall be in accordance with the techniques or conditions described in the literature in this field or in accordance with the product specifications. For reagents or instruments whose manufacturers are not specified, they are all conventional products that can be obtained by purchasing through regular channels.

[0059] In the specific embodiments of the present invention, the cigarette specification parameters are from China Tobacco Jiangsu Industrial Co., Ltd.

[0060] Example 1

[0061] This embodiment provides a method for recommending the replacement of single-leaf cigarettes in a tobacco blend based on a bipartite graph path selection aggregation strategy. A certain cigarette tobacco blend XHM is selected as the implementation object, and the specific steps are as follows:

[0062] (1) Obtain the formula information of each cigarette tobacco blend, including the grade, type, and price of each tobacco blend; the quantitative indicators of the characteristic parameters of each single-leaf cigarette, such as fragrance type, flue gas, aroma, taste, dosage, etc., and construct the global tobacco blend data set as shown in Table 1 below. Due to space limitations, only the tobacco blend data of XHM is shown in the table, where single-leaf cigarette A is the one to be replaced. The characteristic vector of the tobacco blend includes [tobacco blend grade, tobacco blend type, price]; the characteristic vector of the single-leaf cigarette includes [fragrance type, degree, concentration, strength, aroma, aroma quantity, permeability, off-flavor, irritation, dryness, aftertaste]. In addition, the dosage percentage of each single-leaf cigarette in the tobacco blend is used as additional data to participate in subsequent calculations.

[0063] (2) According to the affiliation relationship between each tobacco blend and different single-leaf cigarettes, construct a graph structure G=(V, E), where V={v l 1, v l 2, …, v l n; v c 1, v c 2, …, v c m} refers to the nodes in the graph, and E<,> refers to the edges between the nodes. According to the connection relationship between different nodes, calculate the adjacency matrix A[i, j] of the graph, as shown in Table 2 below. v l refers to the tobacco blend node, and v c refers to the single-leaf cigarette node. In particular, for the set of tobacco blend nodes v l and the set of single-leaf cigarette nodes v c , since there are no edges connecting them internally, the values of A[v l i, v l j] and A[v c i, v c j] in the adjacency matrix are both 0, and only A[v l i, v c j]=*, when the node v l i and vc The edge between j and <v l i, v c When > exists, *=1, otherwise *=0. Where v l 1 refers to the XHM leaf group, v c 1 to v c 5 refers to the single-stem cigarettes A, B, C, D, E. Therefore, in the table <v l 1, v c 1 to 5> has a value of 1. Then, according to the adjacency matrix table, draw a schematic diagram of the bipartite graph structure of the leaf group-single-stem cigarette, as shown in Figure 1 , where v l and v c are the leaf group node and the single-stem cigarette node respectively. The connection between the two is the edge E. There is no edge inside the leaf group node and the single-stem cigarette node. The nodes as a whole present two relatively independent parts, that is, the bipartite graph structure model.

[0064] Table 1

[0065]

[0066]

[0067] Table 2

[0068] <![CDATA[v l 1]]> <![CDATA[v l 2]]> <![CDATA[v l 3]]> …… <![CDATA[v l n]]> <![CDATA[v c 1]]> <![CDATA[v c 2]]> …… <![CDATA[v c m]]> <![CDATA[v l 1]]> 0 0 0 0 0 * * * * <![CDATA[v l 2]]> 0 0 0 0 0 * * * * <![CDATA[v l 3]]> 0 0 0 0 0 * * * * …… 0 0 0 0 0 * * * * <![CDATA[v l n]]> 0 0 0 0 0 * * * * <![CDATA[v c 1]]> 1 * * * * 0 0 0 0 <![CDATA[v c 2]]> 1 * * * * 0 0 0 0 …… * * * * * 0 0 0 0 <![CDATA[v c m]]> * * * * * 0 0 0 0

[0069] (3) Calculate the fusion feature vectors of each leaf group where || represents vector concatenation, fl j =[leaf group grade, leaf group type, price] is the leaf group attribute information; fc i =[aroma type, degree, concentration, strength, aroma, aroma quantity, permeability, off-flavor, irritation, dryness, aftertaste] is the single-stem cigarette characteristic parameter; k is the number of single-stem cigarette types in the leaf group j, refers to the weighted sum of the usage percentages of the single-stem cigarette parameters in the leaf group. Before calculation, convert the descriptive parameters such as [leaf group grade, leaf group type, aroma type] into digital variables. Among them, the leaf group grade is generally divided into Grade A-1 (numbered 1), Grade A-2 (numbered 2), Grade B-1 (numbered 3), Grade B-2 (numbered 4); the leaf group type can be divided into flue-cured type (numbered 1), blended type (numbered 2), flavor type (numbered 3), cigar type (numbered 4); the aroma type can be divided into light aroma type (numbered 1), intermediate aroma type (numbered 3), strong aroma type (numbered 5); taking the XHM leaf group as an example, fl = [1, 1, 800];

[0070] (4) For the fusion feature vector of the leaf group

[0071] Perform a normalization operation. For numerical variable parameters [price, degree, concentration, strength, aroma, aroma quantity, volatility, off-odor, irritation, dryness, aftertaste], the normalization calculation formula is where x is the original value of the variable, x max is the maximum value of the variable, x min is the minimum value of the variable, x norm is the normalized value, with a value range of 0 - 1; for descriptive parameters such as [leaf group grade, leaf group type, flavor type], retain the values. Taking the XHM leaf group as an example, The normalized eigenvector values are [2, 1, 0.8, 1, 0.74, 0.71, 0.69, 0.78, 0.74, 0.73, 0.73, 0.74, 0.72, 0.73]; calculate the fusion feature similarity, using a method that combines the cosine of the angle and the Euclidean distance. For example, the normalized eigenvector values of another adjacent leaf group XDM that uses the same single tobacco as XHM are Then the calculation process of the similarity of the fusion features of the XHM and XDM leaf groups is as follows: For the normalized numerical variable parameters [price, degree, concentration, strength, aroma, aroma quantity, volatility, off-odor, irritation, dryness, aftertaste], use the Euclidean distance to calculate

[0072] For the descriptive parameters [leaf group grade, leaf group type, flavor type] that are converted into digital variables, use the cosine of the angle to calculate: The calculated value of the fusion feature similarity index of the XHM leaf group and the XDM leaf group is Calculate all adjacent leaf groups that use the same single tobacco as XHM, and obtain the three closest leaf groups: XDM (1.172), XCD (1.212), XDD (1.235);

[0073] (5) Construct a global relevant subgraph for the single tobacco A to be replaced, including XHM and its adjacent leaf groups, that is, composed of the XHM, XDM, XCD, XDD leaf groups and relevant single tobacco nodes, as Figure 2, starting from the single cigarette A node and ending with the single cigarette nodes of 12 fourth-order adjacent leaf groups, explore the forward path sequences of this subgraph. There are 12 forward path sequences, namely A->XHM->B->XCD->[1,2,3,4]; A->XHM->C->XDM->[5,6,7,8]; A->XHM->D->XDD->[9,10,11,12];

[0074] (6) Calculate the information fusion feature vector of all forward path sequences in the global correlation subgraph of single cigarette A. The calculation formula is: where fc t is the feature vector of the single cigarette to be replaced; fc ki is the feature vector of the alternative available single cigarette node i in the adjacent leaf group k; is the fusion feature of the leaf group where the single cigarette to be replaced is located; is the fusion feature vector of the adjacent leaf group k, and w k is the weighted value. The calculation formula is: where is the fusion feature vector of the leaf group where the single cigarette to be replaced is located obtained from the previous step calculation and the fusion feature vector of the adjacent leaf group k The similarity between them; taking the path sequence A->XHM->B->XCD->1 as an example:

[0075] where fc A That is, the feature vector of single cigarette A, which is [1,0.62,0.67,0.64,0.67,0.63,0.61,0.67,0.67,0.67,0.67]; is the fusion feature vector of leaf group XHM, which is [2,1,0.8,1,0.74,0.71,0.69,0.78,0.74,0.73,0.73,0.74,0.72,0.73]; for leaf group XCD: w k =1.172 / (1.172 + 1.212 + 1.235)=0.324; fc1 = [1,0.62,0.68,0.69,0.66,0.67,0.69,0.74,0.71,0.78,0.64]; so fpath A→1= [1, 0.62, 0.67, 0.64, 0.67, 0.63, 0.61, 0.67, 0.67, 0.67, 0.67, 2.97, 1.32, 1.03, 1.32, 0.95, 0.94, 0.9, 1.02, 0.98, 0.96, 0.96, 0.97, 0.95, 0.97, 1, 0.62, 0.68, 0.69, 0.66, 0.67, 0.69, 0.74, 0.71, 0.78, 0.64];

[0076] (7) Calculate the information fusion feature vector of all reverse path sequences in the global correlation subgraph of single cigarette A. The calculation formula is Take the path sequence 1 -> XCD -> B -> XHM -> A as an example: fc1 = [1, 0.62, 0.68, 0.69, 0.66, 0.67, 0.69, 0.74, 0.71, 0.78, 0.64];

[0077] fc A = [1, 0.62, 0.67, 0.64, 0.67, 0.63, 0.61, 0.67, 0.67, 0.67, 0.67]; w k = 0.324, so fpath 1→A = [1, 0.62, 0.68, 0.69, 0.66, 0.67, 0.69, 0.74, 0.71, 0.78, 0.64, 3.65, 1.32, 0.96, 1.32, 0.88, 0.93, 0.87, 0.99, 0.97, 0.95, 0.95, 0.96, 0.95, 0.97, 1, 0.62, 0.67, 0.64, 0.67, 0.63, 0.61, 0.67, 0.67, 0.67, 0.67];

[0078] (8) Calculate the similarity of the positive and reverse global fusion features between all alternative single cigarette nodes and the target single cigarette node to be replaced using the Manhattan distance where, x i and y i are the respective feature parameters in fpath t→ki and fpath ki→t The smaller this value is, the more similar the two are. Take fpath A→1=[1,0.62,0.67,0.64,0.67,0.63,0.61,0.67,0.67,0.67,0.67,2.97,1.32,1.03,1.32,0.95,0.94,0.9,1.02,0.98,0.96,0.96,0.97,0.95,0.97,1,0.62,0.68,0.69,0.66,0.67,0.69,0.74,0.71,0.78,0.64] and fpath 1→A =[1,0.62,0.68,0.69,0.66,0.67,0.69,0.74,0.71,0.78,0.64,3.65,1.32,0.96,1.32,0.88,0.93,0.87,0.99,0.97,0.95,0.95,0.96,0.95,0.97,1,0.62,0.67,0.64,0.67,0.63,0.61,0.67,0.67,0.67,0.67], for example, Sim(fpath A→1 ,fpath 1→A )=1.81; for the global fusion features of all forward and reverse path sequences, their similarities are calculated and it is found that Sim(fpath A→3 ,fpath 3→A )=1.24;Sim(fpath A→7 ,fpath 7→A )=1.33;Sim(fpath A→10 ,fpath 10→A )=1.59; they are the three most similar path sequences, so single-ingredient cigarettes 3, 7, and 10 are recommended to replace single-ingredient cigarette A.

[0079] Example 2

[0080] In order to evaluate the effect of the single-ingredient cigarette replacement method proposed in the present invention, this example uses the traditional sensory evaluation experience method of the tobacco industry to evaluate and analyze the implementation results in Example 1. The specific steps are as follows:

[0081] (1) Single-ingredient cigarettes 3, 7, and 10 were used to replace single-ingredient cigarette A in the XHM leaf group, respectively, and three groups of experimental cigarette samples were prepared according to the original formulation ratio. Then, cigarette products with the original formulation of the XHM leaf group were used as the control group.

[0082] (2) Four professional cigarette tasters were recruited to form an expert group to conduct four rounds of sensory evaluation on the three groups of experimental cigarette samples and the control group of cigarette products. The evaluation indicators included: flavor type, degree, concentration, strength, aroma, aroma volume, permeability, impurities, irritation, dryness, and aftertaste.

[0083] (3) In the same round, if the deviation of the scores given by experts for any sensory evaluation index of the same cigarette sample exceeds ±0.2, then re-smoking scoring is carried out for this cigarette sample until the scores of the sensory evaluation indexes meet the deviation range.

[0084] (4) Calculate the average of the scores of each expert in the same round to obtain the final sensory evaluation index parameters of the single-flavor cigarette 3 experimental group, the single-flavor cigarette 7 experimental group, the single-flavor cigarette 10 experimental group, and the original XHM cigarette product control group respectively.

[0085] (5) Quantitatively evaluate the similarity of the sensory evaluation parameters of each alternative experimental group and the XHM control group. The calculation formula is: Where xi is the sensory evaluation parameter of the experimental group and yi is the sensory evaluation parameter of the control group. The final results are shown in Table 3 below. The sensory evaluation similarities of the three alternative leaf groups and the original leaf group formula are all higher than 96%, indicating that the leaf group single-flavor cigarette substitution method of the present invention can effectively carry out single-flavor cigarette substitution.

[0086] Table 3

[0087]

[0088] In summary, the present invention designs a leaf group single-flavor cigarette substitution method based on the bipartite graph path selection aggregation strategy. By constructing a bipartite graph data model of the leaf group formula - single-flavor cigarette information and adopting the path selection aggregation strategy, the single-flavor cigarette to be substituted is globally expressed in all leaf group formula data from two dimensions: the attributes of the leaf group it belongs to and its own characteristics. It not only targets the characteristic parameters of the target single-flavor cigarette, but also completes the recommendation of the substituted single-flavor cigarette from the perspective of leaf group adaptation, which can avoid the uncertainty of traditional single-flavor cigarette recommendation relying on artificial sensory evaluation and improve the scientificity and efficiency of the single-flavor cigarette substitution selection work of the leaf group formula.

[0089] The applicant declares that the above is only the specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Those skilled in the art should understand that any changes or substitutions that can be easily thought of by those skilled in the art within the technical scope disclosed by the present invention fall within the protection scope and the disclosure scope of the present invention.

Claims

1. A method for substituting single-leaf tobacco in a cigarette leaf group formula, characterized in that, The method includes the following steps: (1) Obtain the formula information of each cigarette leaf group, where the information includes the attribute information of the cigarette leaf group formula and the characteristic information of all single-cut tobacco in it; (2) Based on a bipartite graph, model each cigarette leaf group formula and the single-cut tobacco information in it. The nodes in the bipartite graph are the formula information of each cigarette leaf group or the single-cut tobacco in it; (3) Fuse the attribute information of the target cigarette leaf group formula in the bipartite graph and the characteristic information of all single-cut tobacco in it as the fusion feature of the target cigarette leaf group formula; (4) Select the cigarette leaf group formula nodes adjacent to the target cigarette leaf group formula and perform step (3) to obtain the fusion features of the adjacent cigarette leaf group formulas, where the adjacent cigarette leaf group formulas refer to all cigarette leaf group formula nodes that use the same single-cut tobacco as the target cigarette leaf group formula; (5) Calculate the similarity of the fusion features between the adjacent cigarette leaf group formulas and the target cigarette leaf group formula, and select the top K adjacent cigarette leaf group formulas with the highest similarity to the target cigarette leaf group formula; (6) Select the single-cut tobacco nodes in the K adjacent cigarette leaf group formulas obtained in step (5) as the alternative available single-cut tobacco; (7) Construct a global correlation subgraph of the single-cut tobacco node to be replaced. Starting from the single-cut tobacco node to be replaced and ending at the alternative single-cut tobacco node, explore all path sequences in the global correlation subgraph respectively for global feature fusion; (8) Starting from all alternative available single-cut tobacco nodes in the global correlation subgraph constructed in step (7) and ending at the single-cut tobacco node to be replaced, explore all paths in the subgraph and then perform global feature fusion respectively; (9) Calculate the similarity of the positive and negative global fusion features between all alternative single-cut tobacco nodes and the target single-cut tobacco node to be replaced, and select the top M single-cut tobacco with the highest similarity to the target single-cut tobacco node to be replaced as the final recommended alternative single-cut tobacco.

2. The substitution method of single cut tobacco in the cut tobacco blend formula of a cigarette according to claim 1, characterized in that, The information in step (1) specifically includes the grade, type, price of the cigarette leaf group formula, and the aroma, fragrance type, smoke, aroma, and taste characteristic parameters of each single-cut tobacco; Preferably, step (3) specifically includes: starting from the single-cut tobacco to be replaced in the target cigarette leaf group formula, fusing the attribute information of the cigarette leaf group formula of its first-order neighbors and second-order neighbors in the bipartite graph structure and the characteristic information of all single-cut tobacco in it; Preferably, step (5) specifically includes: using a similarity measurement calculation method, comparing the distance between the fusion feature vectors of all adjacent cigarette leaf group formulas and the target cigarette leaf group formula as the similarity evaluation index, and selecting the top K cigarette leaf group formulas with the closest distance to the target cigarette leaf group formula; Preferably, step (7) specifically includes: starting from the single-cut tobacco node to be replaced, its first-order neighbors are the cigarette leaf group formula to which it belongs, the second-order neighbors are the other single-cut tobacco in the cigarette leaf group formula to which it belongs, the third-order neighbors are the K adjacent cigarette leaf group formulas, and the fourth-order neighbors are the single-cut tobacco in the K adjacent cigarette leaf group formulas. The above four parts constitute the global correlation subgraph of the single-cut tobacco node to be replaced. Then, starting from the single-cut tobacco node to be replaced and ending at the fourth-order node, explore all paths in the global subgraph and perform feature fusion respectively; Preferably, step (8) specifically includes: in the global relevant sub-graph obtained in step (7), starting from the 4th-order adjacent node to the single tobacco leaf node to be replaced as the end point, reverse-exploring all paths in the global sub-graph to perform global feature fusion respectively; Preferably, step (9) specifically includes: adopting a similarity measurement calculation method, comparing the distances between the global fusion feature vectors of all alternative single tobacco leaves and the target single tobacco leaf node to be replaced as the similarity evaluation index, and selecting the M single tobacco leaves with the closest distance to the target cigarette leaf group formula.

3. The method for substituting single cut tobacco in the cut tobacco blend formula of a cigarette according to claim 1 or 2, characterized in that, Step (1) further includes obtaining the grade, type, price of the cigarette leaf group formula and the characteristic vectors of the flavor, smoke, aroma, and taste of each single tobacco leaf, and constructing a data set, wherein the characteristic vectors of the cigarette leaf group formula include the cigarette leaf group formula grade, cigarette leaf group formula type and price, and the characteristic vectors of the single tobacco leaf include flavor, degree, concentration, strength, aroma, aroma amount, permeability, miscellaneous gas, irritation, dryness and aftertaste; Preferably, step (1) further includes obtaining the usage percentage of each single tobacco leaf in the cigarette leaf group formula.

4. The substitution method of single-cut tobacco in the cut tobacco blend formula of a cigarette according to any one of claims 1-3, characterized in that, Step (2) specifically includes the following steps: (2.1) Construct a graph structure G=(V, E) according to the respective cigarette leaf group formulas and the belonging relationships of different single-component cigarettes, where V = {v l 1, v l 2, …, v l n; v c 1, v c 2, …, v c m} represents the nodes in the graph, that is, all cigarette leaf group formulas and single-component cigarettes; E represents the edges between the nodes, that is, the connections between the cigarette leaf group formulas and single-component cigarettes. If there is a belonging relationship between the two, then there is a corresponding edge between the cigarette leaf group formula node v l and the single-component cigarette node v c , otherwise there is none; (2.2) Calculate the adjacency matrix of the graph Among them, <v l i, v c j> refers to the edge between nodes v l i and v c j. That is, if there is an edge between the cigarette leaf group formula node v l i and the single-flavor cigarette node v c j, then the value of their adjacency matrix is 1, otherwise it is 0. Based on this, the two-dimensional adjacency matrix A[i, j] can be obtained; Among them, for the cigarette leaf group formula node set v l and the single tobacco leaf node set v c , since there is no edge connection between them, the values of A[v l i, v l j] and A[v c i, v c j] in the adjacency matrix are both 0. Only between the cigarette leaf group formula node and the single tobacco leaf node, A[v i i, v l j] = *, where * = 1 when <v l i, v c j> exists, otherwise * = 0; (2.3) According to the results of steps (2.1)-(2.2), draw a bipartite graph structure diagram of the cigarette blend-single tobacco leaves, where v l and v c are the cigarette blend nodes and single tobacco leaf nodes respectively, and the connection between the two is the edge E. There is no edge within the cigarette blend nodes and single tobacco leaf nodes. The nodes are presented as two relatively independent parts as a whole, namely the bipartite graph model.

5. The method for substituting single-stem tobacco in the cut tobacco blend of a cigarette according to any one of claims 1-4, characterized in that, Step (3) specifically includes the following steps: (3.1) Let the characteristic vector of single cut tobacco i be fc i = [aroma type, degree, concentration, strength, aroma, aroma quantity, permeability, off-flavor, irritation, dryness, aftertaste]; the percentage of single cut tobacco used in the cigarette leaf blend formula is w i , and the characteristic vector of the cigarette leaf blend formula is fl j = [cigarette leaf blend formula grade, cigarette leaf blend formula type, price range]; the integrated characteristic vector of the cigarette leaf blend formula is (3.2) Cigarette leaf group formula fusion feature vector The calculation formula is: Where || represents vector splicing, k is the number of single tobacco types in cigarette leaf group formula j, It means to perform weighted summation of the usage percentage of each single tobacco parameter in the cigarette leaf group formula. The obtained cigarette leaf group formula fusion feature vector can be expressed as: Among them, 'aroma type', 'degree', 'concentration','strength', 'aroma', 'aroma quantity', 'volatility', 'off-flavor', 'irritation', 'dryness', 'aftertaste' are the weighted fusion values of each single tobacco index parameter in the cigarette leaf group formula.

6. The substitution method of single tobacco in the cut tobacco blend formula of cigarettes according to any one of claims 1-5, characterized in that, Step (5) specifically includes the following steps: (5.1) For the fusion feature vector of the cigarette leaf group formula Perform a normalization operation. For the numerical variable parameters price, degree, concentration, strength, aroma, aroma quantity, permeability, off-odor, irritation, dryness, and aftertaste, the normalization calculation formula is where x is the original value of the variable, x max is the maximum value of the variable, x min is the minimum value of the variable, x norm is the normalized value, and the value range is 0 - 1; convert the cigarette leaf group formula grade, cigarette leaf group formula type, and "aroma type" into digital variables. The cigarette leaf group formula grade is divided into Grade A-1, Grade A-2, Grade B-1, and Grade B-2; the cigarette leaf group formula type is divided into flue-cured type, blended type, flavor type, and cigar type; the aroma type is divided into light aroma type, intermediate aroma type, and strong aroma type; and then participate in the subsequent similarity calculation; (5.2) Calculate the fusion feature similarity. Adopting a method combining the cosine of the included angle and the Euclidean distance, for the normalized numerical variable parameters price, degree', concentration', strength', aroma', aroma amount', permeability', miscellaneous gas', irritation', dryness' and aftertaste', the Euclidean distance is used for calculation: For the descriptive parameters cigarette leaf group formula grade, cigarette leaf group formula type and flavor' converted into digital variables, the cosine of the included angle is used for calculation: where x i and y i are the respective characteristic parameters in different eigenvectors. The calculation formula for the final fusion feature similarity of the cigarette leaf group formula is The smaller this value is, the more similar the fusion features of the two cigarette leaf group formulas are; (5.3) Calculate all neighboring cigarette leaf group formulas and the target cigarette leaf group formula to calculate the Sim Select the K neighboring cigarette leaf group formulas with the highest similarity among them.

7. The method for substituting single-cut tobacco in the cut tobacco blend formula of a cigarette according to any one of claims 1-6, characterized in that Step (6) specifically includes: taking the target single tobacco leaf as the starting point, the first-order adjacent node as the belonging cigarette leaf group formula, the second-order adjacent node as other single tobacco leaves in the same cigarette leaf group formula, the third-order adjacent node as the adjacent cigarette leaf group formula node, and the fourth-order adjacent node as the single tobacco leaf of the adjacent cigarette leaf group formula, and the single tobacco leaf of the adjacent cigarette leaf group formula is the alternative available single tobacco leaf.

8. The substitution method of single cut tobacco in the cut tobacco blend formula of cigarettes according to any one of claims 1-7, characterized in that, Step (7) specifically includes the following steps: (7.1) Construct a global relevant sub-graph of the single tobacco leaf node to be replaced, and the global relevant sub-graph includes the target single tobacco leaf in step (6) as the starting point, the first-order adjacent node, the second-order adjacent node, the third-order adjacent node and the fourth-order adjacent node; (7.2) For the global correlation subgraph described in step (7.1), explore all path sequences from the target single cut tobacco node to the K fourth-order adjacent cut tobacco leaf group formula single cut tobacco nodes The calculation formula for the information fusion feature vector of each path sequence is: where fc t is the feature vector of the single cut tobacco to be replaced; fc ki is the feature vector of the alternative available single cut tobacco node i in the adjacent cut tobacco leaf group formula k; is the fusion feature of the cut tobacco leaf group formula where the single cut tobacco to be replaced is located; is the fusion feature vector of the adjacent cut tobacco leaf group formula k, w k is the weighted value, and the calculation formula is: where is the fusion feature vector of the cut tobacco leaf group formula where the single cut tobacco to be replaced is located, calculated using step (5.2) and the fusion feature vector of the adjacent cut tobacco leaf group formula k is the similarity between them.

9. The method for substituting single-stem tobacco in the cut tobacco blend of a cigarette according to any one of claims 1-8, characterized in that, Step (8) specifically includes: Explore the path sequence from the alternative available single tobacco leaf node to the single tobacco leaf node to be replaced The calculation formula for the information fusion feature vector of each path is as follows: where fc t is the feature vector of the single tobacco leaf to be replaced; fc ki is the feature vector of the alternative available single tobacco leaf node i in the adjacent cigarette leaf group formula k; is the fusion feature of the cigarette leaf group formula where the single tobacco leaf to be replaced is located; is the fusion feature vector of the adjacent cigarette leaf group formula k, w k is the weighting value, and the calculation formula is: where is the fusion feature vector of the cigarette leaf group formula where the single tobacco leaf to be replaced is located calculated by step (5.2) and the fusion feature vector of the adjacent cigarette leaf group formula k the similarity between them; Preferably, step (9) specifically includes: obtaining the forward and reverse path fusion feature vectors fpath from the single tobacco leaf node t to be replaced to the alternative available single tobacco leaf node ki according to steps (7) and (8). t→ki and fpath ki→t After that, the Manhattan distance is used to calculate the similarity between the two, and the calculation formula is: where x i and y i are the respective feature parameters in fpath t→ki and fpath ki→t respectively. The larger the similarity value, the more similar the two are. Select the M alternative single tobacco leaves ki with the highest similarity as the final recommended alternative single tobacco leaves.

10. Application of the method for replacing single tobacco leaves in the cigarette leaf group formula according to any one of claims 1-9 in designing a cigarette leaf group formula.

Images